Tag: ai

AI Price War Breaks Out: Meta Unveils Paid AI Model For First Time, Will Be “Among Most Affordable Options”

Shortly after a leaked Meta memo revealed the company was planning on putting an AI chip into production in September as it looks to double computing capacity to 14Gigawatts, the company also unveiled a version of its most advanced artificial intelligence model, Muse Spark 1.1, that includes a new paid tier for developers, marking the first time Meta has charged businesses for access to its models and providing a new revenue stream. It’ll be “among the most affordable options“ on the market, Zuckerberg said in a Bloomberg interview ahead of the release.

“Since this is not an open source model, this is I think the first time that we’re doing a real serious API,” Zuckerberg said, referring to the application programming interface used to access Meta’s AI. “And the pricing is going to be very aggressive and attractive” he added indicating that Meta hopes to capture market share by undercutting its competitors, offering the new model at 25% of the cost of top models from OpenAI and Anthropic.

The new model’s biggest improvement is in its agentic capabilities, the Meta CEO told Bloomberg, and according to benchmarks the model does indeed appear to be in line with the competition.

He hopes to piggyback on the latest craze in AI development this year, which a month ago saw Goldman forecast that agentic AI use will lead to a massive 120 quadrillion monthly tokens being used by 2030.

Agents are the big theme of AI this year, with the label applied to systems that can complete multistep tasks on behalf of a user. Zuckerberg described Muse Spark 1.1 as having “state-of-the-art or very close to it” agentic reasoning and tool use. The model is also greatly improved when it comes to coding and Meta employees are using it internally to build products and features for various apps, he added. 

Meta will also introduce a new Meta Model API system, which will be used to collect fees from developers. Its API pricing is roughly 25% of the cost advertised by other top models from OpenAI and Anthropic, according to Bloomberg. Developers will be able to use Meta’s model for free, but only up to a point; they’ll be required to pay for access after reaching a certain token threshold, Zuckerberg said. 

Which means that legacy frontier models will now have to worry about domestic cheap alternatives, especially after xAI also released an agentic and coding model yesterday which will have to grab market share, in addition to much cheaper Chinese models.

“The pricing from some of the other labs is very extreme and has very high margins,” Zuckerberg said, underscoring that his strategy is to get Meta’s technology in front of as many people as possible. “We think that there’s a real ability to be able to offer frontier or very high-level intelligence at a much more affordable cost.”

Zuckerberg, 42, is spending aggressively to keep pace with rivals like OpenAI and Alphabet in a race to achieve what he calls superintelligence, or AI that can perform tasks better than humans. Meta has committed hundreds of billions of dollars to building the infrastructure necessary to develop superintelligence, including data centers and expensive AI chips. The company announced a new $10 billion data center investment in Canada as well as a new image-generation model just this week.

Keep reading

Leaked Meta Memo Shows AI Capacity Doubling To 14 Gigawatts

Meta shares fell 4.3% at Thursday’s open after Reuters reported the contents of an internal memo laying out the next phase of the company’s AI infrastructure program.

The stock has clawed back part of the loss through the morning but stayed solidly red while the tape digested the same question it has been chewing on for nine days: is Meta the hyperscaler that just started exercising capex discipline, or the one that just committed to doubling?

Three things to note from today’s news. The first is silicon. Iris, Meta’s in-house AI accelerator and one of four planned MTIA generations unveiled in March, enters production at TSMC in September after clearing bug validation in six weeks with no major issues – an unusually clean result for a program that has stumbled for more than half a decade. Broadcom is the design partner under an agreement extended through 2029, and Meta plans to ship a new chip roughly every six months through 2027, against an industry norm of annual-or-slower cadences. The chips are meant to augment, not replace, externally sourced GPUs – Meta separately holds a multiyear agreement with AMD covering up to six gigawatts of Instinct accelerators – but the internal memo is very blunt about why the program matters – as adopting the latest external GPUs at Meta’s scale “has been a heavy lift, and it has cost us time.”

The second is scale. Meta plans to deploy seven gigawatts of computing infrastructure this year and to double overall capacity to fourteen gigawatts in 2027, with 2026 spending running as high as $145 billion – the very top of the range guided in April, and a meaningful slice of the more than $700 billion Big Tech is projected to pour into AI this year.

The third is supply. The memo reveals long-term contracts for memory from Samsung, flash storage from Sandisk and fiber-optic equipment from Sumitomo Electric – multi-year lock-ins struck in the middle of a memory shortage severe enough to be raising consumer hardware prices.

On its face the chip news is bullish: faster, cheaper, more independent compute is exactly what a company spending $145 billion a year should want. But the market has spent the past week and a half developing a very specific allergy, and the memo triggered it.

When Bloomberg reported at the start of the month that Meta was standing up a cloud business – internally, Meta Compute – to sell surplus capacity and token-metered API access to outsiders, the stock ripped nearly 9% higher in a session while CoreWeave and Nebius fell double digits. We suggested this might be a potential first crack in the AI capex boom: hoarding compute stops making sense the moment you admit you have extra, and if management appears willing to monetize idle infrastructure, the market reads capital discipline and pays for it. Days later, leaked town-hall remarks in which Zuckerberg conceded that agent development “hasn’t accelerated in the way we expected” knocked the stock back down – the July 2 drop that Thursday’s open just eclipsed.

Against that backdrop, a memo describing a doubling of capacity, a six-month silicon cadence and years of locked-in component supply looks rather – undisciplined when it comes to capex. Companies do not sign multi-year memory contracts in the middle of a shortage in order to stand still. As we noted earlier this month – the pivot to rewarding CapEx cutters – has, for now, been a driving force: up on plans to sell capacity, down on plans to double it, with the same infrastructure underneath both headlines.

Keep reading

How Flock Cameras Wrongly Tracked Me for Days Over ‘Stolen’ Plates and Sent Police After Me

Are you armed?!” the police officer screamed. “Get out of the car!”

On an otherwise normal Sunday afternoon in late June, I’d decided to take the $155,000 Range Rover I was testing that week out to run some errands with my wife. Little did I know that choice would complete a technological chain linking surveillance cameras, AI, and law enforcement that led to me and my wife being surrounded by police, hands on their guns, in a Kohl’s parking lot in suburban Minnesota.

After dropping off our Amazon returns, we’d just gotten back in the Range Rover and reversed maybe two feet out of the spot when four cop cars came flying out of nowhere and boxed us in. The officers jumped out and started shouting. It’s a situation that can quickly and frequently turn bad, so as unprepared as I was, I followed their orders, got out with my hands up, and tried to figure out what the hell was happening.

Eventually, after a tense hour, I did. The Plymouth Police Department had been tracking me for days using Flock license plate cameras, waiting for the right moment to strike, because they thought I’d stolen the Range Rover. And the reason I was ID’d as a dangerous car thief was a simple data error made 2,000 miles away in California, creating an edge case within an edge case that Flock’s AI camera network was unable to handle.

We now live in a surveillance state where cameras mounted on stoplights are tracking our cars, our devices, our pets, and even us. This is just the beginning; next, these cameras could be put in motion using our kids’ school buses. Whether you’ve actually stolen a car or are just rolling down the road having done nothing wrong, like me, once these systems have you in their crosshairs, there’s pretty much only one way it can go. Welcome to the future. It’s scary out there.

Keep reading

Deadly bacteria found in major US city’s wastewater system tied to Mark Zuckerberg’s $800m data center

Meta‘s massive AI data center in Wyoming is facing scrutiny after an unexpected contamination incident emerged during construction.

The Mark Zuckerberg-owned company is developing a 715,000sq ft campus in Cheyenne that is set to go online next year, but its contractor has come under fire after city officials traced wastewater containing a rare bacterium to the project.

Known as Cupriavidus gilardii, the naturally occurring bacterium is typically found in soil and water. While harmless to most healthy people, it can cause severe pneumonia, bloodstream and lung infections, and, in rare cases, death among people with weakened immune systems. 

Cheyenne’s Board of Public Utilities (BOPU) said the bacterium was found in wastewater discharged by Goat Systems, a contractor working on Meta’s $800 million data center. 

According to the BOPU, the bacterium was first detected during routine wastewater sampling in late February, but was only announced last Thursday.

Meta said its general contractor, Fortis, began hauling industrial wastewater offsite and that independent testing found no trace of the substance to date.

Officials stressed that it did not contaminate the city’s drinking water, but said it disrupted the municipal reclaimed water system and required months of cleanup. 

However, the city permanently revoked Meta’s authorization to discharge wastewater from its fill-and-flush operations into Cheyenne’s treatment system, where the water is recycled and later used to irrigate parks and other public spaces. 

A Meta spokesman told the Daily Mail: ‘When the board shared that it found a substance in the city’s wastewater – not public drinking water – Fortis immediately stopped discharging industrial wastewater and began hauling it offsite.

‘Fortis also began its own water testing with an independent environmental specialist, which has found no trace of the substance. 

‘Meta is committed to being a good neighbor in Cheyenne, including through the protection of local water resources, and will continue encouraging collaboration between Fortis and the board until this situation is resolved.’

It comes as AI data centers face mounting scrutiny across the US for their enormous demands on local water and power supplies. 

According to Data Center Map, there are nearly 4,500 data centers nationwide, with some facilities consuming as much as 300,000 gallons of water a day, roughly the same amount used by 1,000 households.

Goat Systems LLC is the corporate entity Meta uses for the construction of the center, dubbed Project Cosmo.

Officials said the contaminated wastewater was discharged during a fill-and-flush process used to prepare the data center’s cooling system before it goes online. 

Keep reading

Current State Of Physical AI: Everything You Need To Know

Citi’s Robotics & Physical AI Leadership Conference wrapped up on Tuesday. The annual Citi Research event brings together robotics founders, investors, operators, and industry executives to assess the state of “physical AI.”

Analyst Heath Terry summarized the key takeaways Wednesday morning, painting a picture of the robotics industry moving from proof of concept to commercial deployment, while warning that scaling robots remains challenging.

“Labor shortages, reshoring, and favorable regulatory tailwinds are accelerating enterprise demand, while data scarcity, talent constraints, battery limitations, and high deployment costs remain key friction points,” Terry explained to clients. 

Citi said the winners in physical AI will likely be firms that own proprietary real-world data, solve specific labor bottlenecks and use Robotics-as-a-Service models to reduce upfront costs for customers.

Terry highlighted automation-exposed industrial names including Rockwell Automation, Emerson Electric, Honeywell, Symbotic, Ralliant and Belden as potential beneficiaries.

Humanoids are attracting significant investor interest. Last month, we detailed how readers can invest ahead of a major ramp in humanoid production expected in the coming quarters. Read the report. 

Over the last two years, about $20 billion has been invested in physical AI, with applications spanning warehouses, logistics, trucking, construction, aviation, and defense.

Last week, carmaker BMW revealed that a new upgraded humanoid is walking its factory floors at the Spartanburg plant in South Carolina. 

Keep reading

YouTube defends video that falsely claims Sydney massacre survivor is ‘crisis actor’

A Google executive told an inquiry on Tuesday that a YouTube video that falsely claimed a wounded survivor of an antisemitic massacre in Sydney was a crisis actor blooded with makeup had met the platform’s standards and would remain online.

Google Australia manager Rachel Lord was testifying at a government inquiry into the spread of antisemitism in Australia including an attack by two gunmen on a Sydney Hanukkah celebration in December that left 15 dead.

Lord was questioned about a complaint made by survivor Arsen Ostrovsky about a video posted on YouTube. Ostrovsky was attacked online after an image showing blood streaming from a wound in his head was posted on X two hours after he was shot.

Lord said the decision to allow the video to remain on YouTube had been reviewed at “quite senior levels.”

“We have spent a lot of time thinking about where we draw the line and we continue to re-evaluate where we are doing that,” Lord said.

Richard Lancaster, the lawyer leading the inquiry’s evidence, referred to a transcript of the video to avoid showing the images in public.

Four men appear on split screen saying Ostrovsky’s bleeding head appeared “very crisis actor-ish” and mentioned “makeup.” They also describe him as an “intelligence asset” who had a “degree in theater.”

The video also describes Ostrovsky as a Zionist and claims the massacre was a “false flag operation.” Police allege father and son shooters Sajid and Naveed Akram were inspired by the Islamic State group.

Lancaster told Lord the video remaining online demonstrated a “really serious deficiency” in YouTube’s hate speech guidelines.

Lord replied that she appreciated Lancaster’s “feedback.”

YouTube told Australia’s online safety regulator three days after the massacre that the platform was “focused on ensuring Australians and all users around the world have access to high quality information about the tragic events,” Lord said.

Ostrovsky told the inquiry last month that he had been targeted by online hate, abuse, vilification and AI manipulation since he suffered the minor head wound on Dec. 14.

The inquiry was then shown an AI-generated image of Ostrovsky apparently laughing as someone applied fake blood to his head.

Keep reading

Waymo Robotaxi “Snitches” On Two 15-Year-Olds Drinking & Shooting Orbeez Guns In Bay Area

Two 15-year-old boys were detained in San Mateo Monday afternoon after the Waymo robotaxi they were riding in reported them to police – for drinking alcohol and firing a gel-bead blaster out of the moving car – then pulled itself over so officers could collect them.

Waymo’s remote monitors spotted the behavior on the vehicle’s interior cameras and called the San Mateo Police Department around 2:10 p.m. with the car’s exact location. The company then disabled the vehicle near 20th Avenue and El Camino Real, telling the pair the car was having trouble – a ruse that bought officers time to get into position.

Because the initial report described what looked like a real firearm, police conducted a high-risk stop, approaching with guns drawn and a police dog deployed. No one was hurt. Inside, officers found an Orbeez-style gel blaster – painted over to pass for the real thing – and open alcohol.

The teens cooperated, were detained, and were released to their parents. The case has been forwarded to the San Mateo County District Attorney’s office for review of possible charges, including underage drinking, and police say they plan to pull the Waymo’s interior video.

“Parents do you know where your teens are? @waymo does!” The department wrote on Facebook: “After calling us and stopping the car, we were able to safely remove both subjects and determined they were shooting Orbeez from the car as they sipped on afternoon libations while being chauffeured around town in the driverless vehicle.”

“While there was some ingenuity to this scheme, toy guns, water guns, and BB guns all pose real dangers, especially to an untrained eye… Shooting projectiles at speed can cause real damage. And lest not forget the underage drinking. All bad ideas today for these two. Well, the Waymo might have been the smartest idea yet, because driving impaired would’ve made this so much worse.”

Keep reading

Risk and AI: It’s Tricky

A funny thing happens on the way to understanding risk: we discover it’s tricky. We think we see all the risks, and think we can mitigate or hedge those risks, but by its very nature, risk evades such simplistic filters and metrics. Risk remains hidden, offscreen, invisible, building up out of sight, awaiting a catalyst that’s equally undetectable until it manifests, and after the fact, we look back and ask, why didn’t we see that coming?

Risk is tricky like that. It can lay dormant for decades and then erupt with little warning.

Risk is tricky in other ways. In our hubris, we see the power and might of our technologies, systems and foresight, and reckon these are so robust they will easily survive any tectonic shift, as we’ve planned for emergencies.

But our faith in the might of our civilization is itself a source of risk because the risk of Model Collapse–the breakdown not of a supply chain or technology but of our entire conceptual construct of how the world works–goes unrecognized because our confidence that our model maps the real world is so high that we are incapable of recognizing its drift into hallucination and civilizational psychosis.

In other words, our confidence that our conceptual mythologies are accurately mapping the real world is itself a source of civilizational risk because this confidence makes it inevitable that we do more of what’s failing, as the alternative–recognizing our conceptual models and mythologies are self-serving rationalizations that substitute artifice for realistic appraisals–is conceptually and emotionally impossible.

Put another way: Emperor Norton’s delusions of power and grandeur were harmless as long as he was recognized as delusional. But should Emperor Norton actually be given the power he believed was his to wield, then risk rises accordingly.

Consider the bet being made globally that the current iteration of AI will be 1) immensely profitable (the most important thing in the Universe) and 2) immensely productive (secondary to immensely profitable but necessary as a motivation for everyone to throw trillions of dollars at purveyors of AI). The risk that this bet–and the assumptions that make it not only rational but pressing–is the equivalent of handing Emperor Norton the keys to the kingdom with little evidence he will be a wise leader, is unimaginable in the current model / mythology, and so therefore it doesn’t exist.

The worst that could possibly happen in the current model / mythology is a brief spot of bother in the stock market as euphoric overvaluations come down to Earth, and then the immense profits start flowing and markets rocket higher in a multi-decade Bull Market of AI Productivity.

The possibility that the current iteration of AI is innately incapable of metaphorically boiling away the seas is not on the screen, any more than a stock market crash or social upheaval is on the screen. Yet if the fantasy of vast, unstoppable floods of profits driven by vast increases in productivity fail to materialize on a very short timeline, then both a stock market crash and social upheaval move from “impossible” straight through “unlikely” to “happening now,” leaving everyone who thought they understood risk and were properly hedged against unwelcome change in a state of disbelief and wonderment.

Keep reading

Air Force Engineer Accused Of Cutting Down AI Cameras Becomes Unlikely Hero, Raises Thousands For Legal Defense

A U.S. Air Force engineer charged with allegedly destroying a series of AI-powered license plate surveillance cameras has become an unlikely cause célèbre among privacy advocates, drawing thousands of dollars in donations to help fund his legal defense, according to Yahoo News.

Jeffrey Sovern, a 41-year-old Air Force engineer and mechanic from Virginia, is accused of cutting down multiple Flock Safety license plate reader cameras. He now faces 13 counts of destruction of property, along with six counts each of petit larceny and possession of burglary tools.

The case comes as Flock Safety’s automated license plate reader network continues to spread rapidly across the country. Supporters say the cameras help police solve crimes, while critics argue they create a growing surveillance network that tracks the movements of ordinary Americans and raises serious privacy concerns.

Yahoo News writes that opposition to the systems has intensified in some communities, with vandals reportedly using everything from spray paint and garbage bags to chainsaws to disable or destroy the cameras.

Sovern has made no secret of his views. In a GoFundMe campaign created to cover his legal expenses, he framed the case as a fight over privacy rights.

“My name is Jeff and I appreciate my privacy. I appreciate everyone’s right to privacy, enshrined in the fourth amendment,” Sovern wrote.

He said the criminal case has taken a significant emotional toll on him and those close to him, adding that the encouragement he has received online prompted him to launch the fundraiser.

Originally seeking $8,500, the campaign has gained momentum as news of the case has spread. It has now brought in more than $15,000 from over 400 contributors, far surpassing its initial goal.

In a recent update following a preliminary hearing, Sovern thanked supporters for helping bring attention to the issue.

“Thank you to those that had the time to show support this week!” he wrote. “We have seen a huge uptick in awareness of the system and this case.”

He also urged supporters to continue advocating against what he called an expanding surveillance network, encouraging people to “reach out to the local governments and demand that these systems are taken down.”

Keep reading

Zuckerberg to spend over $10B on “historic” Alberta AI data centre investment, sources say

Meta Platforms, the parent company of Facebook, Instagram and WhatsApp, is behind a massive artificial intelligence data centre planned for Sturgeon County, Alberta, Juno News has confirmed.

This is according to several well-placed sources with direct knowledge of the investment, one that all sources agree will be “historic” in magnitude.

The project is expected to involve roughly $13 billion in total investment, though the final figure could still change as it is unclear if the final proposal has been approved and signed off by Mark Zuckerberg and Meta’s board.

Keep reading